This chapter examines the rapid rise of AI companionship—illustrated by Friend.com’s “always-on” wearable assistant—and asks whether such systems can count as friends or durably relieve loneliness. We define AI companions as LLM-based agents designed to simulate responsive, caring partners (friend, confidant, lover, even deceased relative). While acknowledging user-reported benefits and therapeutic promise, we defend three claims: (1) in the canonical sense of friendship, reciprocal person-level commitment, accountability, and shared vulnerability are absent, so one cannot be friends with conversational AIs; (2) many short-term gains (feeling heard, cared for) arise from design features—anthropomorphic framing, asymmetric dependence, and data-driven attunement—that render the value fragile and potentially deceptive; and (3) at scale, commercial incentives and displacement effects risk deepening isolation, crowding out human ties, and normalizing surveillance-mediated intimacy. We close with design and policy implications aimed at mitigating harm while resisting the substitution of artificial companions for human relationships.